The Market Performance of Tracking Stocks

Transcription

1 The Market Performance of Tracking Stocks Matthew T. BILLETT and Anand M. VIJH * Henry B. Tippie College of Business University of Iowa, Iowa City, IA ABSTRACT Tracking stocks have been a popular form of equity restructuring in recent years. AT&T, Disney, General Motors, Sprint, US West, and many others have issued tracking stocks. While the positive announcement returns of tracking stocks are well documented, an examination of their post-issue market performance is lacking. This paper examines the postissue returns and the subsequent restructuring events through December 2000 by using a comprehensive sample of tracking stocks. We document three key results. First, we find that tracking stocks earn significantly negative buy-and-hold excess returns during a three-year period following the issue date. We also find significantly negative returns surrounding the earnings announcements during this period. This evidence contrasts with the post-issue returns of spinoffs, which are known to be positive, and of carveouts, which are known to be insignificant. Second, contrary to a common justification given to adopt tracking stocks, we find that they do not increase the transparency of firm earnings. Third, we find large positive announcement-period returns to events resulting in the elimination of tracking stock structure. JEL Classification: G12; G14; G24; G34. Keywords: Tracking stocks; Divestitures; Second events; Long-term returns; Event study. Current version: February 2001 Comments welcome * Matt-Billett@uiowa.edu and Anand-Vijh@uiowa.edu. Phone: (319) and (319) The paper has benefited from comments of seminar participants at the Case Western Reserve University, the University of Missouri, and the University of Oklahoma. We wish to thank Jon Garfinkel and Ingo Natusch for helpful suggestions. We are obliged to Mr. Stan Levine of the Thomson Financial Company for providing the First Call analyst forecast data.

2 1 The Market Performance of Tracking Stocks 1. Introduction Tracking stocks are big news. On November 22, 1999, the Wall Street Journal reported that AT&T was moving closer to issuing a tracking stock for its fast-growing wireless unit. AT&T s stock rose by percent that day. This was AT&T s biggest one-day return in a decade, even exceeding the return following the September 1995 announcement that it would spinoff its NCR computer business. Tracking stocks were first introduced in 1984 with the creation of GME shares issued by General Motors, to track the performance of its Electronic Data Systems (EDS) division. General Motors followed with another tracking stock, GMH, to track the performance of its Hughes Aircraft division, but the next company to adopt a tracking stock was not until 1991 when USX Corp. separated its oil business from its steel business. While relatively rare until the mid 1990s, tracking stocks have become an increasingly popular form of restructuring. Many companies have issued or proposed issuing tracking stocks, including AT&T, Disney, DLJ, DuPont, General Motors, Microsoft, R.J. Reynolds, Sprint, TCI, USX Corp., US West, and WorldCom. What are tracking stocks? These are newly-issued stocks; created by distributing a non-taxable stock dividend to existing shareholders, by an initial public offering, or as payment for target shares in a merger. A tracking stock is an equity claim intended to reflect the performance of a certain division of a multi-division firm. We refer to the old stock as the general division (GD) stock and the new stock as the tracking (TR) stock. Tracking stocks differ from spinoffs and carveouts that also divide the cash flows of the old firm. Whereas spinoffs and carveouts divide the old firm into two separate firms with distinct boundaries, tracking stocks leave it as one combined firm for legal purposes. This distinction is important, as many benefits of divestitures are believed to result from the complete separation of non-synergistic businesses. Numerous studies document that increasing firm focus by divestiture is associated with an increase in firm value, regardless of whether the divestiture is accomplished by an asset sale, a spinoff, or

3 2 a carveout. 1 This benefit of divestitures is unachievable with tracking stocks. The complete separation of businesses also removes potential conflicts arising from the division of cash flows. Hass (1996), Logue, Seward, and Walsh (1996), Billett and Mauer (2000), D Souza and Jacob (2000), Elder and Westra (2000), and Zuta (2000) discuss the pros of tracking stocks. First and foremost, firms issuing tracking stocks often argue that the analysts and investors cannot understand the value of disparate businesses and therefore undervalue the combined stock. Creation of quasi pure-play tracking stocks attracts greater analyst coverage (the transparency effect) and increases attention from investors interested in different parts of the firm s business (the clientele effect). Second, tracking stocks help attract and retain top managerial talent, whose compensation can be linked to the market value of their divisions. Third, tracking stocks help in accomplishing mergers with target firms whose shareholders are less willing to exchange their stock for the stock of a large and diversified acquiring firm. Fourth, tracking stocks preserve the internal capital markets of diversified firms. Consistent with these arguments, Logue, Seward, and Walsh (1996), Billett and Mauer (2000), D Souza and Jacob (2000), and Zuta (2000) document mean announcement returns of around three percent, which are of the same order as the announcement returns for spinoffs and carveouts. 2 Billett and Mauer (2000) find a positive relation between the announcement returns and the wealth gains or losses from preserving the internal capital markets of the combined firm. The cons of tracking stocks are discussed by Hass (1996), who argues that these are fictional stocks that may create potential conflicts of interest, arising from disproportionate ownership of the GD and TR stocks by directors and managers and sibling rivalry between shareholders of the two stocks. This paper examines the long-term market performance of this unique form of restructuring that separates the stocks without separating the businesses. We analyze 28 TR and 19 GD stocks that were issued during Our sample includes every tracking stock issued in the U.S. market during this 1 Lang and Stulz (1994), Berger and Ofek (1995, 1999), and Comment and Jarrell (1995) document that focus and firm value are positively related and that firm value increases when focus is improved. John and Ofek (1995), Desai and Jain (1999), and Vijh (1999) find that focus-increasing divestitures are associated with greater wealth gains for samples of asset sales, spinoffs, and careveouts, respectively. 2 See Hite and Owers (1983), Miles and Rosenfeld (1983), and Schipper and Smith (1983) for the announcement returns of spinoffs, and Schipper and Smith (1986) for the announcement returns of carveouts.

4 3 period. We first measure the buy-and-hold excess returns (BHERs) over a one-year period before the announcement date, between the announcement and issue date, and over a three-year period after the issue date. We compute these excess returns by using three different benchmarks: the Center for Research in Security Prices (CRSP) value-weighted market returns, the Media General Financial Services (MGFS) industry returns, and the CRSP size decile returns. We find that GD stocks underperform the market and size decile benchmarks during the preannouncement period. On average, the BHERs are significantly negative. This evidence contrasts tracking stocks with spinoffs and carveouts. Desai and Jain (1999) find that pre-announcement BHERs of spinoffs are insignificant, and Vijh (1999) finds that pre-announcement BHERs of carveouts are significantly positive. The evidence supports a frequent conjecture that tracking stocks are issued in response to poor stock price performance. The industry-adjusted BHERs are insignificant, which suggests that the industry as a whole may be underperforming. The pre-issue BHERs, spanning the period from just before the announcement date to the issue date, are generally insignificant. The post-issue returns are of the greatest interest, as this is when the TR stocks begin their lives. Over a three-year holding period starting with the issue date and ending no later than December 2000, the GD stocks earn BHERs that are negative but insignificant. However, the TR stocks earn BHERs that are negative and statistically significant with respect to all three benchmarks. Nineteen of the 28 TR stocks underperform all three benchmarks, only five outperform all three, and four are in-between. To give some feel for the magnitude of underperformance, the TR stocks earn an average buy-and-hold return of 20.9 percent. Over an average holding period of 2.70 years, this translates into an annual raw return of 7.3 percent. In comparison, the annual market return equals 19.3 percent, the annual industry return equals 16.0 percent, and the annual size decile return equals 14.7 percent. The difference between the annual raw return and the benchmark return ranges between 7.4 and 12.0 percent, which is economically significant. We next examine several subsets of data to test the reasons given to explain the issuance decision. We find that none of the subset BHERs are significantly positive, while many are significantly negative. First, during the post-issue period, the smaller TR stocks earn significantly negative returns, which are

5 4 also significantly less than the returns earned by the bigger TR stocks. It appears that issuing tracking stocks linked to small divisions is not in the shareholders interest. Second, TR stocks that partition a firm into growth and value divisions earn insignificant returns while TR stocks that achieve no such purpose earn significantly negative returns. Third, TR stocks that are issued as part of mergers earn significantly negative returns. Thus, receiving payment in the form of a TR stock instead of the acquirer s old stock does not benefit the target shareholders. Tests of long-term returns are sometimes criticized because they cannot control for all known factors in stock returns. In comparison, tests of short-term event-period returns are free of such criticism. We therefore examine three-day announcement-period returns for all quarterly earnings announcements of GD and TR stocks during the same three-year holding period. Consistent with the evidence on longterm returns, the earnings announcement-period returns of GD stocks are insignificant, but the earnings announcement-period returns of TR stocks are significantly negative. In fact, nearly one-third of the negative BHERs of TR stocks during the three-year holding period can be explained by the negative earnings announcement returns. After examining stock returns, we investigate the common transparency argument given to justify the creation of tracking stocks. Using the First Call analyst forecast data, we examine the forecast errors for the combined stock during the four quarters before announcement and the separated stocks during the four quarters after issue. We find no decrease in the magnitude of forecast errors for GD stocks, but a significant increase for TR stocks. We next examine the dispersion of multiple analyst forecasts for any given quarter, and find a small decrease for GD stocks but a large increase for TR stocks. We finally examine the magnitude of the market reaction to earnings announcements. The market reaction increases for both GD and TR stocks, but the difference is not statistically significant. Overall, these tests provide evidence that is inconsistent with the increased transparency argument. There is no improvement in the transparency of GD stocks from before announcement to after issue, and there may be some deterioration in the transparency of TR stocks. The combined evidence on three-year holding-period returns, earnings announcement-period returns, and the transparency of firm s earnings suggests that TR stocks are poor performers while the GD

6 5 stocks are average performers. This evidence is inconsistent with market efficiency, although we must caution that tracking stocks are relatively new and their population is relatively small. The long-term evidence in our study contrasts with previous studies of tracking stocks that document positive market reaction on the announcement of tracking stocks. It also contrasts tracking stocks with spinoffs and carveouts. Cusatis, Miles, and Woolridge (1993) and Desai and Jain (1999) document that the spinoff subsidiary stocks earn significantly positive long-term excess returns, and Vijh (1999) shows that the carveout subsidiary stocks earn insignificant excess returns. If the poor long-term returns reflect problems that were not foreseen, then we would expect many reversals of the tracking stock structure in favor of the old one-stock structure. We would also expect complete divestitures of tracked businesses by spinoff or selloff. In addition, we would expect positive market reaction following the announcement of such reversals and divestitures. To test these predictions, we investigate subsequent restructuring events of TR stocks through December Eight TR stocks have been retired (or proposed to be retired) in favor of the old onestock structure, two have been sold, and two have been spun off. In addition, two TR stocks have been acquired (or proposed to be acquired) along with their GD stock, one has been partially divested by the GD management, and one has been restructured after acquiring another firm. In most cases, the announcement of restructuring events that lead to the elimination of the tracking stock structure is accompanied by strong positive returns to both the TR and GD shareholders. For example, the eight TR stocks that were exchanged for the GD stock realize an average market-adjusted excess return of 19.0 percent, while the corresponding GD stocks realize 7.4 percent. However, the two stocks often react differently, and in a manner consistent with conflicts of interest between their shareholders as highlighted by the media reports. A study of such media reports on the announcement of restructuring events and surrounding events that lead to the elimination of tracking stock structure suggest that TR stocks often create more problems than they solve. We cite evidence of confusion among the analysts and the shareholders as to what they really own, TR shareholders complaining of unfair treatment in dividing the proceeds of restructuring, cross-liability resulting from maintaining the firm as one legal entity, and so on.

7 6 Hass (1996) and many articles in the press have pointed out these problems. 3 Why the proponents of tracking stocks did not foresee these problems at the time of issuing TR stocks is not clear. Where does that leave us? We are unable to explain the market s preference for tracking stocks as shown by the positive announcement returns. It is possible that tracking stocks are advisable in some cases and not advisable in other cases. However, we are unable to find circumstances under which tracking stocks are advisable and leave this topic to future research. The remaining paper proceeds as follows. Section 2 discusses data and methods. Section 3 examines the market performance with longterm returns and earnings announcement returns. Section 4 examines the transparency argument, and Section 5 examines the subsequent restructuring events. Section 6 concludes. 2. Data and methods 2.1. Sample (universe) of tracking stocks during Our sample of tracking stocks is obtained from Billett and Mauer (2000) and D Souza and Jacob (2000) and includes every tracking stock issued in the U.S. market during It includes 20 issues by 14 firms. However, our sample is smaller than the cumulative sample of the above two papers as we exclude cases where tracking stocks were announced but not issued before December Eight firms issued tracking stocks once during the sample period (Ralston Purina Group, CMS Energy Corp., US West Inc., Inco Ltd., Conectiv Inc., Georgia Pacific Corp., Circuit Stores Inc., and Sprint Corp.). Six firms issued tracking stocks twice (General Motors Corp., USX Marathon Group, Pittston Company, Fletcher Challenge Group, Genzyme Corp., and Tele Communications Inc.). Following 19 of the 20 issues, we can unambiguously identify one GD stock and one or more TR stocks. But, in the singular case of Fletcher Challenge Group, there are four comparable-size stocks after the second issue date of March 25, 1996, which are all classified as TR stocks. The overall sample includes 14 different GD stocks and 22 different TR stocks. However, since each issue date is treated as an independent event, there are 19 GD and 28 TR stock observations in the return measurement experiments. In five cases there is some 3 See Burton (1998), McGough (1999 and 2000), Rowland (1999)), Sloan (1999), Vickers (1999), Adamson (2000), Henry (2000), Lashinsky (2000), Malkiel (2000), Scherreik (2000), and Stires (2000). 4 We also exclude one issue by American Health Properties, because it was a real estate investment trust, a preferred stock, and unavailable on the CRSP database.

8 7 overlap between observations as the second issue occurs less than three years after the first issue. This violates the assumed independence of the second event. However, the overlap averages only 1.12 years. Our sample is small, especially for the long-term returns part of our investigation. But it includes the entire population of tracking stocks. Moreover, our results are both economically and statistically significant. Later, we separately analyze a sample of six GD and eight TR stocks that were issued between January 1999 and April The available data cover a much shorter period for these stocks, so we prefer to analyze them separately. We obtain the announcement dates of tracking stock issues from Billett and Mauer (2000) and the Wall Street Journal and the issue dates by tracing the first trading date of TR stocks on the CRSP files. The actual issue date is likely to be on the previous day, but for convenience of exposition we term the first trading date as the issue date. We obtain the earnings announcement dates from the Compustat quarterly file and Lexis/Nexis, and the announcement dates and details of subsequent restructuring events from the Wall Street Journal and Lexis/Nexis Sample distribution over time and the summary statistics Panel A of Table 1 reports the timing and frequency of firms issuing tracking stocks and the number of tracking stocks issued. While GM first introduced tracking stocks in 1984 and 1985, the next company to adopt a tracking stock structure was USX Corp. in The gap between 1985 and 1991 was primarily due to uncertainty regarding the tax treatment of future tracking stocks. In 1991 USX solved this problem and structured their tracking stock in such a way that it was ruled a tax-free event. USX s example led to the wave of tracking stocks that followed. 5 Table 1 shows that one company issued tracking stocks during 1992 and 1994 and three companies issued tracking stocks during 1993 and each year from 1995 to Panel B of Table 1 reports the sample characteristics. The average tracking stock has a market value of $1.6 billion and ranges from $16 million to $9 billion. The ratio of TR to GD market value averages Krishnaswami and Subramaniam (1999) report that the ratio of subsidiary to parent 5 While the IRS has decided not to issue rulings on tracking stocks, the consensus is that for the time being they will be treated as a tax-free form of restructuring (Natusch (2000)).

9 8 market value averages 0.27 for spinoffs, and Vijh (1999) reports that the ratio of offering value to the parent market value averages for carveouts. 6 Thus, in terms of relative size, tracking stocks are comparable to other forms of equity restructuring. Issuers cite a variety of reasons for creating tracking stocks. We find that 29 percent of tracking stocks are the result of a merger where the TR stock gives target shareholders an issue closely aligned to the performance of their old business. Transparency issues and clientele effects are almost always given to justify the issuance of tracking stock. However, we find that 29 percent of our sample of TR stocks are in the same industry as their GD stocks. Transparency and clientele effects are less likely to be the main motivation in these transactions. Finally, 39 percent of the TR stocks divide the combined stock into a growth stock and a value stock. In these cases the separated stocks may be designed to better serve the perceived investor clienteles. Insider holdings align the interests of the managers and other insiders with the interests of the shareholders. One may argue that proportionate holdings of TR and GD stocks would motivate managers to maximize the market value of the combined firm, and disproportionate holdings would motivate them to increase one stock price at the expense of the other. Table 1 reports insider holdings collected from the first proxy statement available after the issuance of tracking stocks. On average, insiders own 3.1 percent of the TR stocks and 2.8 percent of the GD stocks. The average ratio of TR to GD holdings equals The median is a better indicator, and it equals There is considerable cross-sectional variation, and most firms have insider holdings of GD stock substantially different from insider holdings of TR stock. Although not reported, we found that the ratio of CEO holdings of TR to GD stocks is comparable to the ratio of insider holdings, but less skewed, with an average value of 0.79 and a median value of The returns data and the computation of excess returns We compute excess returns by using three different benchmarks to control for the market, industry, and size effects. The computation requires the stock returns, market returns, industry returns, and size decile returns. We obtain all of these returns except industry returns from the CRSP daily return 6 Strictly speaking, Krishnaswami and Subramaniam (1999) report the average ratio of the spinoff to the sum of spinoff and parent value. However, from this we calculate the ratio of spinoff to parent value.

10 9 files ending in December We proxy market returns by returns on the CRSP value-weighted portfolio of NYSE, AMEX, and NASDAQ stocks (symbol VWRETD). Most studies of long-term returns adjust for industry effects by matching sample stocks with other stocks having the same standard industrial classification (SIC) code on CRSP or Compustat files. A recent paper by Kahle and Walkling (1996) documents that 36 percent of the Compustat and CRSP primary SIC codes disagree at the 2-digit level, 50 percent disagree at the 3-digit level, and 79 percent disagree at the 4-digit level, which questions the effectiveness of this procedure. The potential misclassification problem is more acute for tracking stocks. For example, CRSP assigns the same SIC code of 3711 to GM, GME, and GMH stocks. This code stands for Motor Vehicles and Car Bodies, which is descriptive of GM, but not GME or GMH. We searched for alternate sources of industry classification and found a superior source in MGFS. This is a financial services concern that maintains 215 industry indexes. Their classification agrees with hard copy sources that describe the business lines of TR and GD stocks. In the preceding example, MGFS assigns industry groups of Auto Manufacturers Major to GM, Information Technology Services to GME (now EDS), and Communication Equipment to GMH. The MGFS industry classification and returns data have become an industry standard and are used by numerous financial service and information providers. We use the MGFS industry returns data in our analysis, retrieved from the web-site moneycentral.msn.com (owned by the Microsoft Corporation). The data are available for the entire period of our study. However, there is one limitation, that cash dividends are excluded in computing the index returns. As a result, the true industryadjusted excess returns are likely to be more negative than our reported industry-adjusted excess returns. We calculate that the GD and TR stocks included in our study have average annualized dividend yields of 2.82 and 1.95 percent during the sample period. We compute long-term excess returns by subtracting the buy-and-hold market returns, industry returns, or size-decile returns from the buy-and-hold stock returns over an appropriate holding period. The resulting buy-and-hold excess returns, or BHERs, are an accurate measure of excess returns realized by 7 To be more precise, we use the size decile returns for the relevant exchange (NYSE or NASDAQ) that are included in the CRSP index files. However, for brevity of exposition, we refer to these as the size decile returns.

11 10 the long-term shareholders of TR and GD stocks. The underlying buy-and-hold portfolio strategy is easy to implement and requires no subsequent rebalancing. We compute short-term announcement-period excess returns by subtracting the three-day cumulative market returns from stock returns. The three-day measurement period is centered on the event date, which is either the Compustat announcement date or the Wall Street Journal or Lexis/Nexis publication date. For both long-term and short-term experiments, we compute t-statistics by using the cross-sectional distribution of excess returns. In addition to the returns data, this paper analyzes the First Call analyst forecast data obtained from the Thomson Corporation. We describe this later when we report empirical tests based on these data. 3. Long-term returns and earnings announcement-period returns 3.1. BHERs of the aggregate sample of GD and TR stocks Table 2 presents the BHERs of the GD and TR stocks. We examine the stock price performance over the year prior to announcing the tracking stock, the period between announcement and issue, and the three years following the issue. Over the year prior to announcement, the GD stocks earn an average 0.67 percent raw return, percent market-adjusted return (significant at the one-percent level), percent industry-adjusted return (insignificant), and percent size-adjusted return (significant at the one-percent level). The medians present an almost identical picture. Ten of the 18 raw returns are negative. Thus, one plausible motivation for issuing tracking stocks may be to bolster poor stock price performance. This evidence of negative pre-announcement performance is in stark contrast to other forms of equity restructuring. Desai and Jain (1999) find insignificant BHERs prior to spinoffs. Vijh (1999) finds that the pre-announcement BHERs of carveouts are significantly positive. If firms issue tracking stocks in an attempt to improve stock market performance, then it would be interesting to ask whether this new form of equity restructuring succeeds in improving the GD stock price performance. Table 2 shows that the BHERs of GD stocks over a three-year period starting with the issue date average -4.15, -4.86, and percent with reference to the market, industry, and size-decile benchmarks. All figures are statistically insignificant, and arguably economically insignificant. The BHERs during the period between

12 11 the announcement and issue date are also insignificant, except with reference to the industry benchmark. Both the pre-issue and post-issue BHERs show that the GD stocks no longer underperform after the announcement of tracking stocks. On average, they earn returns that are statistically indistinguishable from their benchmarks. The long-term returns of TR stocks may shed more light on whether these restructurings improve firm performance. Similar to spinoffs, carveouts, and equity issues, it also makes more sense to examine the performance of these newly created stocks. The evidence in Table 2 suggests that TR stocks do not outperform their benchmarks over a three-year period. In fact, they considerably underperform. The mean three-year market-adjusted, industry-adjusted, and size-adjusted BHERs equal , , and percent. The associated t-statistics are -2.17, -2.31, and -1.85, significant at the five-percent level in the first two cases, and ten-percent level in the third case. Following Lyon, Barber, and Tsai (1999), we also calculate skewness-adjusted t-statistics as t sa = n 0.5 [ S + γ S 2 /3 + γ/(6n) ], where n is the sample size, S is the ratio of sample average to standard deviation, and γ is the sample skewness. The skewness-adjusted t-statistics for the market-adjusted, industry-adjusted, and size-adjusted BHERs equal -2.00, -2.35, and -1.63, which compare favorably with the unadjusted t-statistics. This shows that skewness is not a major problem in our sample. To further illustrate the statistical significance, Table 2 shows that the three median BHERs equal , , and percent, significant at the five-percent level in each case (not shown in table). Nineteen of the 28 TR stocks earn negative marketadjusted returns, 21 earn negative industry-adjusted returns, and 21 earn negative size-adjusted returns (significant in each case at the five-percent level or better). In addition to the statistical significance, the economic significance of return differences is quite large. Over the average holding period of 2.70 years, the TR stocks earn an average percent return, which is percent less than the market return. This translates into an average annual return of 100 ( /2.70-1) = 7.29 percent for TR stocks and percent for the market. A similar calculation shows an average annual return of percent for the industry stocks and percent for the size decile stocks. Recall that the industry return may be understated by about the same order as the 1.95

13 12 percent average annual dividend yield of TR stocks in our sample. The TR stocks thus underperform the benchmarks by an average of 7.29-(19.28+( )+14.66)/3 = percent a year over a three-year period after the issue. 8 The evidence on long-term returns shows that the TR stocks are poor performers. This evidence is particularly interesting when contrasted with the announcement-period returns. Previous studies by Logue, Seward, and Walsh (1996), Billett and Mauer (2000), D'Souza and Jacob (2000), Elder and Westra (2000), and Zuta (2000) have documented that the announcement of tracking stocks is associated with an average positive abnormal return of around three percent. These studies interpret the positive announcement-period returns as evidence that tracking stocks benefit shareholders. Our study suggests that, at a minimum, they do not benefit the long-term shareholders of TR stocks Cross-sectional differences in BHERs Table 3 reports the three-year BHERs and the industry classification for each of the GD and TR stocks in our sample. It is interesting to note that in many cases the industry classification for TR stocks is the same as that for GD stocks. Also noteworthy is the cross-sectional variation in returns. To get a better idea of the cross-sectional patterns, Table 4 examines the BHERs of TR stocks by various groups. The last panel of Table 4 lists which TR stocks are included in these groups. Panels A and B of Table 4 examine the relation between size and BHERs. Panel A divides the sample into TR stocks with a market value above the median vs. TR stocks with a market value below the median. The market value for this purpose is calculated in December 2000 dollars by using the market returns between the issue date and December Similarly, Panel B splits the sample above and below the median ratio of TR to GD market values. Both panels make the same point. The larger TR stocks have 8 We also examine the long-term returns by using the Fama and French three-factor model. For every GD and TR stock, we calculate monthly returns over a period that begins with the first month after the issue month and ends thirty six months later. We calculate the monthly portfolio return by averaging across all GD or TR stock returns during a calendar month. The model regresses the difference between the monthly portfolio return and the riskfree return on the following three factors: the difference between the market return and riskfree return, the difference between returns on portfolios of small and big stocks, and the difference between returns on portfolios of high and low book-to-market stocks. The intercept in this model serves as a measure of abnormal returns. In unreported results, we find that the intercept for both the GD and TR regressions is insignificant. This is perhaps not surprising in view of our small sample size. For example, the TR calendar-time portfolio consists of a single stock during onefifth of the sample period, and has no more than 14 stocks during any one-month period. Given the sample and timing characteristics of tracking stocks, we are uncomfortable making any inferences based on these tests.

14 13 insignificant mean and median BHERs while the smaller TR stocks have significantly negative mean and median BHERs. The smaller TR stocks also have BHERs that are statistically different from the larger TR stocks. An interesting question concerns the relationship between the pre-announcement performance of the old firm and the post-issue performance of the track. Panel C examines the BHERs classified by whether the GD stock was a poor performer during the year prior to tracking stock announcement. We find that there is practically no significant relationship between the pre-announcement performance of GD stocks and the post-issue performance of TR stocks. Firms issuing tracking stocks often argue that they appeal to a broader investor clientele or that they enhance the firm transparency. Panel D partitions the sample into tracking stocks that divide the firm into a growth stock and a value stock vs. those that have the TR and the GD both classified as a value stock or both classified as a growth stock. We find that TR stocks that separate the firm into a growth and a value stock do not earn BHERs that are significantly different from zero. However, TR stocks that fail to create such distinctive stocks earn significantly negative BHERs. The difference between these two groups is also significant. While there is no evidence to suggest that TR stocks that distinguish growth and value divisions increase stockholder wealth, there is some evidence to suggest that TR stocks that do not provide such distinction destroy shareholder wealth. Panel E reports the results when the sample is stratified according to whether the TR and GD stocks belong to the same industry or different industries. We find no statistical difference between the two groups, although the returns are significantly negative when the TR and GD stocks belong to different industries. Panel F breaks the sample into TR stocks that were merger related vs. the others. Again, the BHERs of the two sub-samples are statistically indistinguishable, although the industryadjusted and size-adjusted BHERs are significantly negative for the merger-related sub-sample. Panels G uses the insider holdings of the GD and TR to determine whether the TR stock returns are correlated with the incentives created by differential stock holdings. It shows that BHERs of cases where the insiders own a smaller percentage of GD stock than TR stock are statistically indistinguishable from cases where the insiders own a larger percentage of GD stock than TR stock. Panel H divides the

15 14 sample into subsets with insider holdings of TR stock above and below the median. The BHERs of TR stocks with above median insider holdings are insignificantly different from zero, while the BHERs of TR stocks with below median insider holdings are significantly negative. However, the difference between the two subset BHERs is not significant. Finally, although not reported in Table 4, we split the sample by using the CEO holdings in place of the insider holdings. The results are very similar. The combined cross-sectional evidence shows that smaller TR stocks perform worse than larger TR stocks, TR stocks that do not separate the growth and value components of the old stock perform worse than those that do, but that there is no difference in performance based on pre-announcement returns, differential insider holdings, industry affiliations of the GD stock vs. the TR stock, and whether the issue is merger related. It is also remarkable that not one of the 96 subset BHERs in Table 4 is both positive and significantly different from zero Additional evidence on BHERs with recent issues Appendix 1 and Table A.1 present additional evidence on the post-issue performance by using a more recent sample of tracking stocks issued between January 1999 and April The short history of these more recent TR stocks prevents us from merging this sample with our primary sample. However, the short history does not hide their poor performance. The average market-adjusted, industry-adjusted, and size-adjusted BHERs of eight TR stocks issued during this period equal , , and percent, negative but statistically insignificant, perhaps due to the small sample size Long-term earnings announcement-period returns Long-term returns should be driven by long-term earnings performance. If investors are overoptimistic about the prospects of TR stocks at the time of issue, then the earnings announcements should come as unpleasant surprises. Jegadeesh and Titman (1993) use the earnings announcement-period excess returns over a three-year period to support their evidence on long-term excess returns of recent winners, and Jegadeesh (2000) uses a five-year period to support his evidence on long-term excess returns of seasoned equity offerings. This methodology has the added attraction that short-period excess returns are robust to the choice of benchmark returns.

16 15 Table 5 reports an event study of earnings announcement dates for GD stocks during a period of one year before announcement and three years after issue, and for TR stocks during a period of three years after issue. The announcement dates are obtained from Compustat, First Call, or Lexis/Nexis (in that order), and are sometimes unavailable. Earnings announced within the first 63 trading days of tracking stock issue are classified as quarter +1, within 64 to 126 trading days as quarter +2, etc. The excess returns are computed by subtracting the cumulative market returns from the stock returns over a three-day period centered on the earnings announcement date. Panel A of Table 5 shows that GD stocks earn a mean excess return of 0.57 percent (t-statistic 1.20) during the pre-announcement period. This insignificant excess return suggests that, on average, the earnings announcements of GD stocks during this period are not significantly different from expectations. We recall from Table 2 that the industry-adjusted BHERs during this period are also insignificant, but the market-adjusted and size-adjusted BHERs are significantly negative. Once again, it appears that the poor performance based on the latter benchmarks may be common to other stocks in the GD industry, and it may not reflect poor earnings performance of the GD stocks before the announcement of tracking stocks. The excess returns during the post-issue period average an insignificant percent (t-statistic -0.26) for GD stocks, but a significantly negative percent (t-statistic -2.46) for TR stocks. The medians and the fraction of positive returns confirm the significance of mean returns. Panel B of Table 5 shows that, averaged across all earnings announcements in an event quarter, the mean excess return is negative in 8 out of 12 cases. For both GD and TR stocks, the post-issue earnings announcement-period excess returns are consistent with their long-term excess returns. To explain the economic significance of TR stock announcement-period returns, it is important to discuss the choice of benchmark excess returns. The above analysis of Table 5 assumes that ex-ante the mean earnings announcement-period excess return should equal zero. This assumption is questioned by the classic Robicheck and Myers (1966) ship set-sail story. 9 Based on this theory, and using all earnings 9 The Robicheck and Myers (1966) story can be explained as follows. Suppose a ship sets out on a long voyage in search of a fortune. No information reaches the market during the time the ship is in transit, so the expected return on equity claims linked to the payoff from the voyage should be the same as risk-free return. The uncertainty is resolved on the day the ship reaches the port. Assuming that the payoff risk cannot be diversified, the expected return on arrival date should be very high.

17 16 announcements during , Chari, Jagannathan, and Ofer (1988) show that the mean earnings announcement-period excess return is positive. Casual empiricism suggests that the 1990s were a period of stronger price increases, and that some of the price increases for individual stocks occurred around the strong earnings news. We therefore measure the excess returns around all earnings announcements reported on the Compustat database during Excluding only stocks with a market value of less than $10 million or a stock price of less than $3 produces a mean excess return of 0.20 percent. 10 Using this benchmark for earnings surprise leads to an excess return of -( ) = percent for TR stocks, with a t-statistic of /0.94 = It also changes the pre-announcement and post-issue mean excess returns for GD stocks to 0.37 and percent, with t-statistics of 0.78 and An earnings announcement-period excess return of percent per quarter explains a cumulative underperformance of around 100 (1-( ) 11 ) = 9.87 percent over 11 announcements for TR stocks during the average holding period of 2.70 years. An excess return of percent explains percent. This is roughly one-third of the total long-term underperformance. The combined evidence suggests that the TR stocks are poor performers during a three-year period after the issue. 4. Earnings transparency of pre-announcement GD stock vs. post-issue GD and TR stocks One commonly cited reason for issuing tracking stocks is to increase the firm transparency. This reason would imply that the earnings forecasts of the sample firms should be more accurate following the track. In other instances, Krishnaswami and Subramaniam (1999) find an increase in the accuracy of earnings forecasts for a sample of spinoffs, while Gilson, Healy, Noe, and Palepu (1998) find a similar increase for a combined sample of spinoffs, carveouts, and tracking stocks. Unfortunately, Gilson, Healy, Noe, and Palepu do not separate their results by the type of restructuring. Below we measure the impact of tracking stocks on firm transparency by collecting information on the four earnings announcements before the announcement of a track and the four announcements following the issuance. 10 This excess return is not driven by any one year. Over each year from 1990 to 1999, the mean excess return equals -0.03, 0.40, 0.32, 0.10, 0.05, 0.06, 0.28, 0.07, 0.18, and 0.48 percent. Changing the market value and stock price filters to $1 million and $1 increases the ten-year mean excess return to 0.27 percent, and changing filters to $1 billion and $10 increases it to 0.33 percent. 11 This calculation ignores the small standard error of the benchmark excess return of 0.20 percent.

18 17 We examine three measures of firm transparency: the magnitude of earnings forecast errors, the dispersion of earnings forecasts across analysts, and the magnitude of market reaction to actual earnings announcements. We calculate forecast errors for each quarter, defined as the actual earnings minus the average forecast all scaled by the stock price, and examine the standard deviation of forecast errors before announcement and after issue. Table 6 shows that there is no significant difference for GD stocks. However, the evidence for TR stocks is unanimous. In all eight cases where we have adequate data, the standard deviation of forecast errors is larger for TR stocks than for the pre-announcement GD stocks. We next analyze the dispersion of forecasts across analysts. The standard deviation of forecasts should be lower if the track increases the firm transparency. For GD stocks, we find that the average standard deviation of forecasts declines slightly after the track. The difference is small and statistically insignificant, although the number of decreases, 11, is more than the number of increases, 3, at the tenpercent level. For TR stocks the evidence is the opposite. The standard deviation of forecasts for TR stocks is three times as large as that for the pre-announcement GD stocks, and the difference is significant at the five-percent level. The non-parametric evidence is stronger. In all nine cases with adequate data, the TR stock had a larger standard deviation of forecasts than the pre-announcement GD stock. As a final measure of transparency, we examine the market reaction to earnings announcements. The standard deviation of the four market-adjusted earnings announcement returns to the GD stocks is slightly larger after the issue, but the difference is statistically insignificant. The corresponding increase for the TR stocks is bigger, but remains statistically insignificant. The combined evidence of Table 6 suggests that there is no improvement in the firm transparency after the issuance of tracking stocks. There is no systematic decrease in earnings forecast errors, the dispersion of earnings forecasts across analysts, or the earnings surprise as measured by the market reaction to actual earnings announcements. There may even be some deterioration in transparency, especially for the TR stocks. This evidence contrasts tracking stocks with spinoffs that lead to improved transparency of the parent stocks as documented by Krishnaswami and Subramaniam (1999). It appears unlikely that the separation of stocks without the separation of underlying businesses makes the valuation of new GD and TR stocks easier than the valuation of old GD stock.

19 18 5. Subsequent events resulting in the elimination of tracking stocks A number of firms adopting tracking stocks subsequently eliminate them. We examine the motivation and wealth effects of all events that result in the partial or complete elimination of tracking stocks. To identify these events, we search the Wall Street Journal and Lexis/Nexis from the date of TR issuance through December The events fall into one of four categories: formal retirement of the entire tracking stock structure, sale of assets associated with a TR stock resulting in its elimination, spinoff of a TR stock, and other miscellaneous events related to the tracking stock structure. Table 7 shows the market-adjusted excess returns earned by the TR and GD stocks following the announcement of such events. We have two main results. First, on average, the elimination of TR stocks results in positive wealth effects for both the TR and GD stocks. Twelve out of 14 TR stocks realize positive excess returns, and five out of seven GD stocks also realize positive excess returns. Second, for a given firm, there can be substantial differences in the reaction of the TR and GD stocks. Differential price reaction suggests that the tracking stock structure successfully separates a firm into distinct economic entities. However, the differential price reaction also indicates that a tracking stock structure leads to conflicts of interest between the TR and GD shareholders. For example, a common method used to retire TR stock involves exchanging TR shares for GD shares at a pre-specified premium, resulting in a wealth transfer between the two classes of shareholders. We examine these events in detail below. Appendix 2 contains a description of the transactions used to eliminate the tracking stock structure. 12 Panel A of Table 7 shows four companies that announce their intention to retire the TR stock by exchanging it for the GD stock. This group includes CMS Energy, Pittston, Fletcher Challenge, and Inco Ltd., involving eight TR stocks and three GD stocks. The average announcement-period excess return equals percent for the eight TR stocks, significant at the one-percent level. 13 The three GD stocks 12 The TR stock prospectus usually states the methods and conditions under which the firm can retire the TR stock. These methods include: 1. Exchanging GD shares for TR shares, typically at a 15 to 20 percent premium, 2. Selling or liquidating the assets associated with the TR division and paying the net proceeds to TR shareholders, and 3. Spinning off the TR division. Of course, following a common practice in stock transactions, at any time after issue the GD firm can make a tender offer to retire the TR shares. 13 Because the eight TR stocks belong to four firms, one may argue that the events are not independent and therefore the significance level is overstated. To address this problem, we average across all TR stocks for a given firm. The

20 19 earn an excess return of 7.40 percent. These excess returns are quite large, especially compared to the 1.58 percent mean excess return associated with the announcement of adopting a tracking stock structure in Billett and Mauer (2000). 14 We gather information on the motives for eliminating the TR structure. The managers of CMS Energy argued the TR stock had served its purpose and was no longer necessary: While Class G stock helped CMS Energy gain market recognition for our gas utility business when it was first issued in 1995, our gas business has since grown to the point where having a separate tracking stock for our gas utility is no longer useful, said William T. McCormick, Jr., CMS Energy s chairman and chief executive officer. (Source: PR Newswire, September 9, 1999.) Pursuant to this goal, the CPG (TR) shares were exchanged for CMS (GD) shares at a 15 percent premium to the pre-announcement market value. Presumably, this premium came at the expense of GD shareholders. The announcement returns are consistent with this notion. CPG stock price went up by percent after adjusting for the market returns while CMS stock price went down by 1.27 percent. While the reasons given by CMS Energy suggest that the tracking stock structure may have served some useful purpose at some time, the reasons given by the managers of Pittston, Fletcher Challenge, and Inco Ltd. for retiring TR stocks suggest that the tracking stock structure simply caused more problems than it solved. In resolving to dismantle the tracking stock structure, Mr. Kerry Hoggard, the chairman of the board of directors of Fletcher Challenge, stated: It is clear that the Group s capital structure is seen as complex by investors, is perceived to raise governance issues, and has resulted in a significant structural discount being applied to all our stocks. We cannot allow this to continue, and will move as quickly as possible to a full dismantling of the targeted share structure. (Source: Fletcher Challenge press release, December 16, 1999.) This statement is particularly interesting in view of the fact that firms adopting tracking stock structures often cite increased transparency and elimination of a conglomerate value discount as primary motives. combined excess returns equal percent for CMS (unchanged), percent for Pittston, percent for Fletcher, and percent for Inco Ltd. (unchanged). The average of these four composite returns equals percent (t-statistics 4.74), which is significant at the five-percent level, despite only four observations. 14 The mean announcement returns to the creation of only these four dismantled TR stocks average 0.67 percent (in each individual case, 0.81 percent for CMS Energy, 4.55 and 5.46 percent for the first and second announcements by Pittston, 1.60 percent for the second announcement by Fletcher (first is unavailable), and percent for Inco).